Benchmark Almanac

scBench (LatchBio SingleCellBench)

GPT-6 Astra 63.9% leads since September 2026
GPT-6 Astra 63.9% leads since September 2026
Only in system cards

195 verifiable single-cell RNA-seq analysis problems. Each pairs a data snapshot with a task and a deterministic grader.

Anthropic system cards call it LatchBio SingleCellBench with 195 problems, which matches the 195 problems in latchbio/scbench. Only six example problems are public, and the rest are withheld.

Made by
LatchBio
Released
February 2026
Kind
Lab and data work
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last in a system card
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026
27
Entries on its own leaderboard, 26 models we track
7
System cards report it, from 1 lab

GPT-6 Astra leads with 63.9%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

GPT-6 Astra leads with 63.9%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

27 entries. Scroll sideways for the rest.

  1. 63.9 GPT-6 Astra
  2. 63.6 GPT-6.1 Sol
  3. 61.9 Claude Opus 5.5
  4. 61.5 GPT-6 Sol
  5. 60.5 Claude Fable 5.1
  6. 60.3 Claude Sonnet 5.5
  7. 59.8 GPT-5.6 Sol
  8. 59.1 Grok 4.6
  9. 59.1 Gemini 4 Argon
  10. 58.5 Grok 4.7
  11. 58.3 Claude Opus 5
  12. 58.3 GPT-6.1 Terra
  13. 57.1 GPT-5.5
  14. 57.1 Claude Opus 4.8
  15. 55.6 Claude Sonnet 5
  16. 54.7 DeepSeek V4.1 Flash
  17. 53.9 GPT-6 Luna
  18. 53.0 Mistral Large 4
  19. 52.5 Claude Opus 4.7
  20. 50.4 Gemini 3.7 Flash
  21. 50.3 GPT-5.6 Luna
  22. 48.7 GLM-5.3
  23. 47.5 Kimi K3
  24. 41.7 Claude Sonnet 4.6
  25. 36.6 Nemotron 3 Ultra 550B A55B
  26. 34.0 Gemini 3.8 Flash
  27. 22.9 Nemotron 3.5 Lightning 30B-A3B
  1. 1 GPT-6 Astra published v1 63.9%
  2. 2 GPT-6.1 Sol published v1 63.6%
  3. 3 Claude Opus 5.5 published v1; LSVP 61.9%
  4. 4 GPT-6 Sol published v1 61.5%
  5. 5 Claude Fable 5.1 published v1; LSVP 60.5%
  6. 6 Claude Sonnet 5.5 published v1 60.3%
  7. 7 GPT-5.6 Sol published v1 59.8%
  8. 8 Grok 4.6 published v1 59.1%
  9. 9 Gemini 4 Argon published v1 59.1%
  10. 10 Grok 4.7 published v1 58.5%
  11. 11 Claude Opus 5 published v1 58.3%
  12. 12 GPT-6.1 Terra published v1 58.3%
  13. 13 GPT-5.5 published v1 57.1%
  14. 14 Claude Opus 4.8 published v1 57.1%
  15. 15 Claude Sonnet 5 published v1 55.6%
  16. 16 DeepSeek V4.1 Flash published v1 54.7%
  17. 17 GPT-6 Luna published v1 53.9%
  18. 18 Mistral Large 4 published v1 53%
  19. 19 Claude Opus 4.7 published v1 52.5%
  20. 20 Gemini 3.7 Flash published v1 50.4%
  21. 21 GPT-5.6 Luna published v1 50.3%
  22. 22 GLM-5.3 published v1 48.7%
  23. 23 Kimi K3 published v1 47.5%
  24. 24 Claude Sonnet 4.6 published v1 41.7%
  25. 25 Nemotron 3 Ultra 550B A55B published v1 36.6%
  26. 26 Gemini 3.8 Flash published v1 34%
  27. 27 Nemotron 3.5 Lightning 30B-A3B published v1 22.9%

Source: Official scBench (LatchBio SingleCellBench) leaderboard

27 models. Scroll sideways for the rest.

  1. 63.9 GPT-6 Astra
  2. 63.6 GPT-6.1 Sol
  3. 61.9 Claude Fable 5.1
  4. 61.9 Claude Opus 5.5
  5. 61.5 GPT-6 Sol
  6. 60.6 Claude Opus 5
  7. 60.3 Claude Sonnet 5.5
  8. 59.8 GPT-5.6 Sol
  9. 59.1 Grok 4.6
  10. 59.1 Gemini 4 Argon
  11. 58.5 Grok 4.7
  12. 58.3 GPT-6.1 Terra
  13. 58.2 Claude Opus 4.8
  14. 57.1 GPT-5.5
  15. 56.2 Claude Haiku 5.5
  16. 56.0 Claude Sonnet 5
  17. 54.7 DeepSeek V4.1 Flash
  18. 53.9 GPT-6 Luna
  19. 53.0 Mistral Large 4
  20. 52.5 Claude Opus 4.7
  21. 50.4 Gemini 3.7 Flash
  22. 50.3 GPT-5.6 Luna
  23. 48.7 GLM-5.3
  24. 47.5 Kimi K3
  25. 41.7 Claude Sonnet 4.6
  26. 34.0 Gemini 3.8 Flash
  27. 22.9 Nemotron 3.5 Lightning 30B-A3B
  1. 1 GPT-6 Astra published v1 63.9%
  2. 2 GPT-6.1 Sol published v1 63.6%
  3. 3 Claude Fable 5.1 61.9%
  4. 4 Claude Opus 5.5 published v1; LSVP 61.9%
  5. 5 GPT-6 Sol published v1 61.5%
  6. 6 Claude Opus 5 60.6%
  7. 7 Claude Sonnet 5.5 published v1 60.3%
  8. 8 GPT-5.6 Sol published v1 59.8%
  9. 9 Grok 4.6 published v1 59.1%
  10. 10 Gemini 4 Argon published v1 59.1%
  11. 11 Grok 4.7 published v1 58.5%
  12. 12 GPT-6.1 Terra published v1 58.3%
  13. 13 Claude Opus 4.8 58.2%
  14. 14 GPT-5.5 published v1 57.1%
  15. 15 Claude Haiku 5.5 56.2%
  16. 16 Claude Sonnet 5 56%
  17. 17 DeepSeek V4.1 Flash published v1 54.7%
  18. 18 GPT-6 Luna published v1 53.9%
  19. 19 Mistral Large 4 published v1 53%
  20. 20 Claude Opus 4.7 published v1 52.5%
  21. 21 Gemini 3.7 Flash published v1 50.4%
  22. 22 GPT-5.6 Luna published v1 50.3%
  23. 23 GLM-5.3 published v1 48.7%
  24. 24 Kimi K3 published v1 47.5%
  25. 25 Claude Sonnet 4.6 published v1 41.7%
  26. 26 Gemini 3.8 Flash published v1 34%
  27. 27 Nemotron 3.5 Lightning 30B-A3B published v1 22.9%

Sources: Official scBench (LatchBio SingleCellBench) leaderboard and labs’ model cards

#EntryScoreSettingDateSource
1 GPT-6 Astra 63.9% published v1, Pi Sep 2026 Official leaderboard
2 GPT-6.1 Sol 63.6% published v1, Codex Sep 2026 Official leaderboard
3 Claude Opus 5.5 61.9% published v1; LSVP, Pi Sep 2026 Official leaderboard
4 GPT-6 Sol 61.5% published v1, Pi Sep 2026 Official leaderboard
5 Claude Fable 5.1 60.5% published v1; LSVP, Pi Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 60.3% published v1, Pi Sep 2026 Official leaderboard
7 GPT-5.6 Sol 59.8% published v1, Codex Jul 2026 Official leaderboard
8 Grok 4.6 59.1% published v1, Pi Aug 2026 Official leaderboard
9 Gemini 4 Argon 59.1% published v1, Pi Sep 2026 Official leaderboard
10 Grok 4.7 58.5% published v1, Pi Sep 2026 Official leaderboard
11 Claude Opus 5 58.3% published v1, Pi Jul 2026 Official leaderboard
12 GPT-6.1 Terra 58.3% published v1, Pi Oct 2026 Official leaderboard
13 GPT-5.5 57.1% published v1, Pi Apr 2026 Official leaderboard
14 Claude Opus 4.8 57.1% published v1, Pi May 2026 Official leaderboard
15 Claude Sonnet 5 55.6% published v1, Pi Jun 2026 Official leaderboard
16 DeepSeek V4.1 Flash 54.7% published v1, Pi Sep 2026 Official leaderboard
17 GPT-6 Luna 53.9% published v1, Pi Sep 2026 Official leaderboard
18 Mistral Large 4 53% published v1, Pi Oct 2026 Official leaderboard
19 Claude Opus 4.7 52.5% published v1, Claude Code Apr 2026 Official leaderboard
20 Gemini 3.7 Flash 50.4% published v1, Pi Aug 2026 Official leaderboard
21 GPT-5.6 Luna 50.3% published v1, Pi Jul 2026 Official leaderboard
22 GLM-5.3 48.7% published v1, Pi Aug 2026 Official leaderboard
23 Kimi K3 47.5% published v1, Pi Jul 2026 Official leaderboard
24 Claude Sonnet 4.6 41.7% published v1, Claude Code Feb 2026 Official leaderboard
25 Nemotron 3 Ultra 550B A55B 36.6% published v1, Pi Nov 2026 Official leaderboard
26 Gemini 3.8 Flash 34% published v1, Pi Sep 2026 Official leaderboard
27 Nemotron 3.5 Lightning 30B-A3B 22.9% published v1, Pi Aug 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6 Astra 63.9% published v1, Pi Sep 2026 Official leaderboard
2 GPT-6.1 Sol 63.6% published v1, Codex Sep 2026 Official leaderboard
3 Claude Fable 5.1 61.9% Sep 2026 Lab model card
4 Claude Opus 5.5 61.9% published v1; LSVP, Pi Sep 2026 Official leaderboard
5 GPT-6 Sol 61.5% published v1, Pi Sep 2026 Official leaderboard
6 Claude Opus 5 60.6% Jul 2026 Lab model card
7 Claude Sonnet 5.5 60.3% published v1, Pi Sep 2026 Official leaderboard
8 GPT-5.6 Sol 59.8% published v1, Codex Jul 2026 Official leaderboard
9 Grok 4.6 59.1% published v1, Pi Aug 2026 Official leaderboard
10 Gemini 4 Argon 59.1% published v1, Pi Sep 2026 Official leaderboard
11 Grok 4.7 58.5% published v1, Pi Sep 2026 Official leaderboard
12 GPT-6.1 Terra 58.3% published v1, Pi Oct 2026 Official leaderboard
13 Claude Opus 4.8 58.2% May 2026 Lab model card
14 GPT-5.5 57.1% published v1, Pi Apr 2026 Official leaderboard
15 Claude Haiku 5.5 56.2% Oct 2026 Lab model card
16 Claude Sonnet 5 56% Jun 2026 Lab model card
17 DeepSeek V4.1 Flash 54.7% published v1, Pi Sep 2026 Official leaderboard
18 GPT-6 Luna 53.9% published v1, Pi Sep 2026 Official leaderboard
19 Mistral Large 4 53% published v1, Pi Oct 2026 Official leaderboard
20 Claude Opus 4.7 52.5% published v1, Claude Code Apr 2026 Official leaderboard
21 Gemini 3.7 Flash 50.4% published v1, Pi Aug 2026 Official leaderboard
22 GPT-5.6 Luna 50.3% published v1, Pi Jul 2026 Official leaderboard
23 GLM-5.3 48.7% published v1, Pi Aug 2026 Official leaderboard
24 Kimi K3 47.5% published v1, Pi Jul 2026 Official leaderboard
25 Claude Sonnet 4.6 41.7% published v1, Claude Code Feb 2026 Official leaderboard
26 Gemini 3.8 Flash 34% published v1, Pi Sep 2026 Official leaderboard
27 Nemotron 3.5 Lightning 30B-A3B 22.9% published v1, Pi Aug 2026 Official leaderboard

Labs report scores for 7 models in their cards. Switch the scores above to rank them together.

6 models have held number one on the official leaderboard since its release in February 2026 8 models have held number one with claims included since its release in February 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic OpenAI
Own leaderboard 6 leaders
GPT-6 Astra 63.9%
Lab claims 3 leaders
Claude Fable 5.1 61.9%

Sources: Official scBench (LatchBio SingleCellBench) leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Opus 4.8 Claude Opus 4.7 Claude Sonnet 4.6 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra GPT-5.5

Sources: Official scBench (LatchBio SingleCellBench) leaderboard and labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

7 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 7, 2026 Claude Haiku 5.5 system card 56.2 LatchBio Bioinformatics, SingleCellBench, max effort, safeguards off (8.13.1, CB-1 evaluation in 2.2.2) Bio and chem risk
Sep 28, 2026 Claude Sonnet 5.5 system card 59.1 LatchBio SingleCellBench (8.17.2, Figure 8.17.A) Capabilities
Sep 22, 2026 Claude Opus 5.5 system card 61.2 LatchBio Bioinformatics, SingleCellBench, 195 problems, max effort (8.17.2) Capabilities
Sep 1, 2026 Claude Fable 5.1 system card 61.9 LatchBio SingleCellBench, Mythos 5.1 Other
Jul 24, 2026 Claude Opus 5 system card 60.6 LatchBio Bioinformatics, SingleCellBench, 195 problems (8.17.2) Capabilities
Jun 30, 2026 Claude Sonnet 5 system card 56 LatchBio SingleCellBench, 195 problems, score 0.56 (8.14.2, Figure 8.14.6.A) Capabilities
May 28, 2026 Claude Opus 4.8 system card 58.2 LatchBio Bioinformatics, SingleCellBench, no extended thinking (8.16.3) Capabilities

The lab's number comes out ahead in 4 of 6 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
Claude Opus 5 60.6% LatchBio Bioinformatics, SingleCellBench, 195 problems (8.17.2) 58.3% Leaderboard, published v1, Pi +2.3 pts
Claude Fable 5.1 61.9% LatchBio SingleCellBench, Mythos 5.1 60.5% Leaderboard, published v1; LSVP, Pi +1.4 pts
Claude Sonnet 5.5 59.1% LatchBio SingleCellBench (8.17.2, Figure 8.17.A) 60.3% Leaderboard, published v1, Pi −1.2 pts
Claude Opus 4.8 58.2% LatchBio Bioinformatics, SingleCellBench, no extended thinking (8.16.3) 57.1% Leaderboard, published v1, Pi +1.1 pts
Claude Opus 5.5 61.2% LatchBio Bioinformatics, SingleCellBench, 195 problems, max effort (8.17.2) 61.9% Leaderboard, published v1; LSVP, Pi −0.7 pts
Claude Sonnet 5 56% LatchBio SingleCellBench, 195 problems, score 0.56 (8.14.2, Figure 8.14.6.A) 55.6% Leaderboard, published v1, Pi +0.4 pts

More in Science and reasoning

All 92 in Science and reasoning
  • GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics. Gemini 4 Carbon 96.5%
  • Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
  • FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics. Intern-S2-397B 87%
  • BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
  • HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels. Gemini 3 Pro 78.3%
  • LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%